SK Group Chairman Chey Tae-won on SK Hynix's massive memory expansion — full interview

Chairman Chey. Thank you so much for having us at SK Hynix here in Korea.

I wanted to start out by asking you, you know, you are in charge of a lot more than just this company, but how important is memory within SK Hynix even though you have more than 100 subsidiaries?

Yes, this is AI era.

So people need a lot of memory.

So does the AI.

So right now, what I really believe AI is a kind of small child, four year old only, but well, as they grow, they may need much more functions and they need a lot of experience and more knowledge. Then you have to store some those kind of

knowledge. Then you have to store some those kind of knowledge in somewhere like a human.

Human need as well, in a child you have some small brain and there is not much memories.

But if you become an adult, then you had a lot of experience. You had a lot of memories and wisdom.

experience. You had a lot of memories and wisdom.

So I think that the AI is almost the same thing.

So that's why we need that lot of memory and that memory chips in AI side.

So, but, among our subsidiary, so, well, the Hynix, Hynix is the biggest subsidiary right now.

So well, that's no wonder.

But we, for the AI, we need energies and networks too. So in that sense that our own

networks too. So in that sense that our own portfolio, we have energy company and network companies call that the telecommunication company right now, but, well, they are trying to make that AI data center in the future.

So we have some kind of optimized at our own the portfolio, what I'm saying.

So, yes, but you are right.

Right now, SK Hynix is the most important piece of the portfolio.

And it wasn't always a semiconductor company.

Your beginnings were in textiles.

I got to see the museum yesterday.

When and how did you decide, and was it you who decided to make this a semiconductor company?

Well, yes, it was 2012.

So what I really decided we needed some other portfolios because, well, we already had some portfolio for the energy side and the chemical side and telecommunications.

But what I really needed some more the global business. So also in the digital era at the time as

business. So also in the digital era at the time as well, what I really see that I, I don't see that the AI at the time, but I really believe that digital society also needed a lot of memories.

So you have to actually put in a lot of knowledge and data and put it into somewhere.

So, well, that's naturally the storage of memory chips.

I just go for that to those companies.

And yeah, this might be the future.

So I made a very risky bet at the time.

But yeah, right now as well actually, one of the biggest and successful bets in my life.

You're making another really big bet now, $720 billion to build it out.

How do you tell people that the long term, sustainable business can sustain that kind of investment?

Well, if you see that the demand forecast, like, people will probably be using more AI agents.

So in the future is AI agent.

What I really see compared to 2025.

And I think that 77 tenfold in a within the five years.

So that much AI agent probably it will be the whole world. So that also requires that a lot of

whole world. So that also requires that a lot of memories. So what I really see is that the memory

memories. So what I really see is that the memory increase in the total volume is within ten years, I guess almost five times current the production capacity. So demand is enormous.

capacity. So demand is enormous.

So what I really believe, in order to meet the world, those kind of demand and we have to actually expand very quickly. So that's what I really decide and why we

quickly. So that's what I really decide and why we should expand our supply capacity.

Right here we have a wafer with a signature on it.

Oh yes.

So that's Jensen's? Yes.

Jensen's? Yes.

Is he one of your friends?

And who are some of your other friends in the AI world?

Oh, well, mostly the hyperscalers.

The Micarosoft Satya's and Google's Sundar and well, Mark and Meta and well, you mentioned everybody needed our customers and mostly they are my friend.

And, well, I occasionally meet them.

How about TSMC?

Oh, of course.

Yeah, C.C. Wei I just met him about a month ago.

So I visited Taiwan and we had dinner together.

So we exchanged a lot of information and the forecasting in the future.

So yeah, he also believed that the AI is a real technology, and we have actually expanded our capacity a lot.

And not only well, even though there is a lot of memory TSMC capacity as a limited, then well, there is not much GPU and vice versa.

So even though TSMC capacity is a lot, but there is not enough HBM, then well, there is no use.

So yeah, I mean.

We have to actually make partnership together and working together. So that's the kind of relationship

working together. So that's the kind of relationship with TSMC too.

Right? They're producing your base die.

Correct. How do you get enough capacity on their manufacturing lines? I mean, everybody is lining up to

manufacturing lines? I mean, everybody is lining up to try to get that capacity.

Well, he knows that the important pieces because without them and there is no AI production, especially in the GPU side.

So no matter how he had that, well the demand is overwhelming, but I believe that he's going to make that our base die. Of course.

die. Of course.

And back to Jensen.

How important is Nvidia as a customer?

Well, it's the most important customer right now. So it's also developed the product in and the new

now. So it's also developed the product in and the new product every year.

So that kind of speed is what I always admire those things. And we also push that new chips and new

things. And we also push that new chips and new technology along alongside with him.

So that's one of the really, really good companies, really admirable, companies and in the world right now.

So I still believe that in the future.

So, they are dominant and they actually had a lot of technologies, not only just GPU side.

And he tried to make the AI ecosystems with others together. So not only his own company,

together. So not only his own company, but he actually made a lot of different type of partners. And he's trying to make that ecosystem

partners. And he's trying to make that ecosystem together. That's the most important matter.

together. That's the most important matter.

Yeah. Does it create risk to have so much of your business with one customer, Nvidia?

Without Nvidia, AI ecosystem is not there.

So actually he created a lot of the solution for the AI infrastructure.

Without his production, so I have to actually support that as much as I can do it.

That's the kind of important matter.

But well, he's not the only the customer.

So I always had some some of the other customers like Google and Microsoft and other hyperscalers, but some of them, they make the chips.

But right now, scale is Jensen's company is largest. So well, that's actually,

largest. So well, that's actually, I'm not really dependent on just one customer, but yeah, I just want to solve for the largest customers in the world right now.

Are any of these customers helping you with this build out? Can they do anything to help you ramp capacity fast

out? Can they do anything to help you ramp capacity fast enough to help them fill their demand, help you fill their demand?

They're always asking more chips.

So that's the one thing.

And I'm trying to bring some solutions, but right now, demand is kind of explosive.

And especially it just occurred that the last year, if we want to expand the capacity, so we need a lot of lead time, at least 4 to 5 years.

So probably next year is the hardest year because all of the memory companies are not actually prepared to expand their capacity yet, but still demand is growing.

I think that the demand is growing twice this year actually. Yeah.

actually. Yeah.

All my customers are asking almost double their demand.

So they ask for more chips in the next years, but yeah, supply is not there.

So that's kind of, I don't know how to say, it's like a war between everybody wants more memory chips. Without that they cannot produce their AI

chips. Without that they cannot produce their AI computing and AI chips.

And also one of the, one of the bad things is the prices went up too fast.

So I think that there is a problem with the PC manufacturers and the smartphone manufacturer, consumer electronics and even automakers has some difficulty to find that the chips and well, some of them, even though their chip is

too expensive. So one of my job is that I want to defend

too expensive. So one of my job is that I want to defend all the memory market.

And I'm trying to actually expand that as much as I can and as fast as I can.

So yeah, memory prices have gone up 40% - 50%. Do you anticipate seeing that continue?

50%. Do you anticipate seeing that continue?

I mean, Apple, one of your customers, has had to raise prices on their flagship products.

Are we going to see that continue?

Well, that's what we call chipflation.

The price of the chip is too high and even Apple, they have to raise the price.

And so that actually, increased all the prices to society. So that's, that's not really good.

society. So that's, that's not really good.

So I don't want that, but it just happened.

So, I don't have any solution in a short term.

Well that's why actually I put in huge money scheduled to expand the fab.

So I already announced that within five years we're going to double up our capacity.

But most of the customers, that's not enough.

So we need more.

So that's, that's the kind of problems right now.

But physically, some things are not possible. So, so that's the one of the

possible. So, so that's the one of the problem. That's what we call the AI bottlenecks.

problem. That's what we call the AI bottlenecks.

Yes. So right now as a memory chip is AI bottleneck.

HBM or DRAM is about three quarters of your revenue.

One solution that others are exploring is trying to architect chips that need other things, aside from DRAM to try to do AI, does that create any risk?

No. Technology is just evolution.

Well, it will be any time.

So DRAM is just based on things.

And with the DRAM, we actually stack up and make an HBM. And we might add that one of the solution for the CXL type, and there is a various type

of the technology which can actually support the AI data centers. So in the future there are a few options.

centers. So in the future there are a few options.

So but I think that our company has actually full support at every possible way to expand our the memory capacity. And well, our final target is that how

capacity. And well, our final target is that how we can actually lower down the token cost for the AI.

And right now, what I really believe is a token price is too high.

So what we have to actually push down that more.

So otherwise people have to spend too much money for the tokens.

One analyst told me that memory companies in Korea historically have put a lot of focus on on CapEx to try to get market dominance.

Sometimes, you know, at at risk of when the bust happens, you've overbuilt.

What do you say?

Yeah. We always carefully doing that and but we just announced that the huge expansion plan.

But that's only the condition and the demand is there. So we actually, every time we're going to

there. So we actually, every time we're going to expand and we will check if demand is there.

And our customer made some long term agreement with us.

And we also had a various programs to how we can actually share those kind of risks together.

What kind of programs?

Well, we can make that some fab joint venture.

I can actually do, what we call memory as a service. So it's like not only just selling the chips.

service. So it's like not only just selling the chips.

But as to whether there are very different types of programs, and while we might share that some some risk on that. And you know, we don't want the

that. And you know, we don't want the oversupply, but what we don't also really want undersupply either.

So, so we always check.

And right now, another big thing is the HBM. People needed a lot of AI accelerators in these

the HBM. People needed a lot of AI accelerators in these days. And probably every country,

days. And probably every country, every society wants to have that over those things.

And in order to get there, we actually produced a lot of HBM. But the problem is that HBM,

of HBM. But the problem is that HBM, once you expand those kind of HBM, it actually consumes more than four times of the normal DRAM. So they actually eat up the space of

normal DRAM. So they actually eat up the space of the wafers. That's the kind of problems.

the wafers. That's the kind of problems. But well, what can you do?

So what we have to do it.

And that's why actually we produced a lot of wafer, the capacity. So but I don't really think that,

the capacity. So but I don't really think that, well, you said that you are right.

There is some risk on the overall supply, but, we already noticed that and we carefully confirm that all the demand is there in the future.

How have you seen the length of contracts or the terms of contracts change because of this?

Number one is that that's not just come from us.

Actually, the customer wants those kind of long term agreements because they have to secure that the memory side. And if customer wants and will provide those

side. And if customer wants and will provide those things and well, if they want to only money making debut, then it doesn't really help us that right now there is a shortage of demand.

And I don't want it to have that we were fixing some price and demand.

Yeah, well, those kind of long term agreement, but I wanted to have some comfort zone to especially our customers. So we yeah, we are offering the long

our customers. So we yeah, we are offering the long term agreement and we're checking that's one things.

And we got to secure that kind of demand.

And every year we have to check it.

And we're going to extend that those long term agreements year by year.

So that's one thing.

And we can do it together.

So yeah, let's just check and don't don't make that over investment or under investment.

So I heard that the hotels around Korea have been packed with tech moguls coming here to try to sign these long term contracts.

Is that true?

Right. Right. But yeah, as I told you, it's probably next year is the worst year for the memory shortage.

So there's a lot of people wants to buy that.

Try to find them, find the chips.

But wow, our supply capacity is limited right now.

So that's why whenever I meet our customers, I'm really sorry about this, but if everybody wants all the chips, then I cannot just provide them automatically but I'm trying to do my best.

So that's the only thing.

But, yeah, next year is a kind of the worst year for the chip shortage wise.

2023 was a hard time for you.

You kept going. 1997, that was also, you know, the company just neared bankruptcy.

Mhm. This time, what's going to be different.

This time?

I don't really believe that this time is different, but one thing different is that, well, we have to deal with the new new things, new baby, the AI.

So AI has actually changed our society.

Also AI consumes a lot of resources, as you know, and they're going to add up a lot of energies, the powers and the materials, the silicons, even water.

So also it just created the intelligence.

Not like old days, the old days.

We produced just product, but yeah, people can use it to this.

Intelligence is different.

So we can be much more smarter, happier. So also there is some danger.

happier. So also there is some danger.

I know that, but we have to carefully watch out for that, that kind of dangers.

But, in general, purpose of the AI as a could actually help that, relieve from the human side is labor. But AI actually makes us that there is not

is labor. But AI actually makes us that there is not much difference between the people to people.

So that actually give us that much more relief from the, the tension of the whole society.

That's what I really believe.

That's actually the benefit of the AI side.

So, we don't have to hard work.

And we also heard that there were no difference in people. Then, while we might have,

people. Then, while we might have, there's more chance to we make it a happy society.

Speaking of power and water, you've mentioned that those resources are more important when choosing where to build than geography, right? Does the U.S.

right? Does the U.S.

have enough resources to support an HBM for production facility?

Whenever as the ecosystems for the, the the memory chips.

If you see that the world that, nice big the memory fab.

But behind the scenes, there is a almost 600, 700 different companies to support their material, chemicals and services.

Otherwise they cannot really survive.

So that's one of the biggest problem.

Every country wants to build that the the memory fab in their territory.

So but if there is a water power and land and also the the problem is that the ecosystem.

So if I can bring that ecosystem together, then well, then then I'm going to build that anywhere. So anywhere in U.S.,

anywhere. So anywhere in U.S.,

including U.S. and anywhere in the world.

Our study team is really looking at where is a real possible opportunity and how fast we can build that new fab. And how we can relieve this shortage of memory.

fab. And how we can relieve this shortage of memory.

So you have the site coming to Indiana in the U.S.

SK Hynix has denied reports that you are acquiring another project from another partner.

But the U.S. government, the U.S.

Commerce secretary, just in July, has put the pressure on for you and Samsung to bring a full memory fab to the U.S.

What are your plans?

Well, as I told you that we've been studying.

So I cannot just announce that yes, we're going to build a fab in somewhere.

But where? And we have to actually decide and the insight that, which, when and how and what we need much more the details.

And we also needed a lot of contracts together.

But we are working on it.

So not only just U.S.

The whole world.

And I'm looking for some expansion.

I'm willing to do it, but the problem is that I really have to find the right place yet.

So yeah, to tell you the truth.

And yeah, we've been studying more than a month.

Studying the U.S.?

U.S. also study somewhere else, and we also study together.

Are you getting calls from the U.S.

government or customers saying build in the U.S.?

Build a fab in the U.S.?

Well U.S. Customer wants to build at the fab on U.S. soil.

And the government?

I actually haven't actually talked with that directly with the government yet.

So but this is kind of business.

So if customer wants it and um, yeah, I'm really want to pursue that this kind of opportunity together.

What is your advantage over Samsung and Micron?

Oh, well, I can say a lot of different things.

But the Hynix as you know that as it's under the chapter 11 companies, they've been there more than